01 · What breaks
Design for the failure modes, not only the happy path.
FLN data looks simple until the second assessment round. Most errors appear when a tool changes, children miss a round or a level is read as a verdict on the child.
Changing assessment versions without mapping
A new reading passage or a revised maths task in year two can shift results on its own. Store the tool version on every assessment and keep a written mapping between versions. Without it, a rise in story-level readers may say more about the passage than about the children.
Reporting only children with an endline
Children who migrate, leave or are absent on endline day are often the ones who learned least. Dropping them from the base inflates gain. Count them as attrition, show that number beside every gain figure, and never fill a missing endline from a midline score or a teacher's estimate.
Treating level as a fixed label
A child placed at letter level in June may be reading words by September. If the level field is typed over, or the group never changes, facilitators keep teaching to the old level. Store each assessment as its own dated record and calculate the current level from the latest one.
Using automated scoring without facilitator review
App-based scoring is quick, but a misheard answer or a child too shy to speak can be scored as a failure. Let the facilitator review and correct scores before they are locked, log who changed what, and keep the raw response wherever the tool allows it.
NIPUN Bharat, the national foundational literacy and numeracy mission launched in July 2021 under Samagra Shiksha, aims for every child to reach FLN by the end of Grade 3 by 2026-27. Many programmes use ASER Centre's free reading tool, which places a child at the highest level they manage comfortably: letter, word, paragraph or story. If you align with either, record which framework and which tool each result comes from.
02 · Minimum viable model
Build the model around a child whose level changes during the year.
The central record is the learner's movement through levels, not a mark sheet. The guardian link carries legal weight too: India's Digital Personal Data Protection Act, 2023 requires verifiable consent of a parent or lawful guardian before a child's personal data is processed. Most of the Act's duties apply from around May 2027 (as of October 2026), but record consent before the baseline now. The Theory of Change to data model guide shows how these records become outcome indicators.
| Record | Key fields | Entered by | When |
|---|---|---|---|
| Learner | Programme ID, PEN or APAAR ID if shared, date of birth, guardian, consent date | Field coordinator | At enrolment |
| Assessment | Round (baseline, midline, endline), domain, tool and version, level reached, assessor, date | Facilitator or assessor | Each round and each short check |
| Back-check | Second assessor, level found, size of disagreement | Mentor | A sample at baseline and endline |
| Group membership | Group, start date, end date, reason for the move | Facilitator confirms | After each assessment |
| Session attendance | Session date, present or absent | Facilitator | Same day |
| Support action | Concern, action agreed, owner, due date, closed date | Mentor | When non-progress is flagged |
03 · Implementation sequence
Move through reviewable gates.
Each step leaves the programme with something it can check at the next assessment round, so mistakes show up within a term rather than in the funder report.
Agree skill progression and assessment rules
Write down the reading and maths levels, the tool and version for each, the assessment windows and who may assess. Decide what moving up means, for example one level or more. Use the same tool at baseline, midline and endline within a year, and record any change before the round begins.
Capture a trustworthy baseline
Assess each child one to one within the first two weeks. A mentor re-checks about one child in ten with a different assessor, and a large disagreement raises an alert before groups are formed. Treat that sample size as a starting point the programme sets for itself, not a sector standard.
Group learners without freezing labels
Let the system propose groups from the latest level and the facilitator confirm or change them, with a reason. Balance group sizes so one facilitator is not left with every beginner. Re-group after each short check, and keep dated membership so attendance and progress stay with the right group.
Record attendance and short checks
Mark attendance on the day, on a phone that works offline where the centre has weak signal. Add short fortnightly checks of a few items instead of waiting for midline. They tell the facilitator who is stuck while there is still time in the year to respond.
Review non-progress and absence together
A child with no gain and low attendance needs a different response from one who attends every session and still does not move. Put both facts on the same monthly review list. Low exposure usually calls for a home visit; no progress despite good attendance may need a closer look at language or learning needs.
Report gain with attrition and missingness
Report gain only for children with a baseline and an endline on mapped tools, split by starting level, and put the attrition count and its reasons on the same page. A single figure such as "number of students improved" hides both the base and the starting point, so it should never be the only number a funder sees.
The figures below are invented to show the layout. Each row compares children who started at the same level, and the 15 children with no endline stay visible instead of disappearing from the base.
| Starting reading level | Children at baseline | Assessed at endline | Moved up one level or more | No endline (attrition) |
|---|---|---|---|---|
| Beginner | 40 | 34 | 22 of 34 | 6 |
| Letter | 60 | 55 | 41 of 55 | 5 |
| Word | 50 | 47 | 30 of 47 | 3 |
| Paragraph | 30 | 29 | 21 of 29 | 1 |
04 · Evidence
Measure whether the work became clearer.
Four checks tell a programme lead whether the data can carry a learning claim: how many children have matched records, how gain differs by starting level, whether sessions and assessments actually happened, and whether flagged children received support. The education sector playbook places these alongside the wider programme picture.
05 · Responsible automation
Let systems prepare. Keep consequence with people.
Systems can support
- Proposing groups from the latest assessed level
- Flagging disagreements between first and second assessors
- Listing children with low attendance and no gain
- Calculating matched gain and attrition by starting level
People should decide
- Assigning remedial support
- Interpreting disability or language factors
- Overriding an assessment
- Communicating progress to families
06 · Common questions
Practical questions teams ask.
Questions programme leads, MEL teams and facilitators raise when FLN assessment moves off paper.
What should foundational literacy and numeracy tracking software do?
It should record each child's reading and maths level on a named tool, link that level to the child's group and attendance, and show movement between assessment rounds. It also needs to flag children who are not progressing so a mentor can act, and to report gain on matched records with attrition shown. A tool that only stores marks by class cannot tell you whether a child moved from word to paragraph reading.
What records and data fields does an FLN tracking system need?
At minimum, a learner record with guardian and consent details, an assessment record for each round with tool, version, level and assessor, dated group membership, session attendance and a log of support actions. Add a back-check record when a different assessor repeats the assessment for some children. The current level should be calculated from the latest assessment, not kept in a field that someone types over.
Who should assess children and who should review the results?
Trained facilitators or assessors usually run the one-to-one assessment, and a mentor or field coordinator re-checks a sample with a different assessor. The facilitator confirms the suggested group. The MEL lead owns tool versions, mappings and the gain calculation, and a programme manager decides on extra support for flagged children. Lock scores only after the facilitator has reviewed them.
What are the most common mistakes in FLN tracking?
Four come up again and again. Tools change between years without a mapping, so gains stop being comparable. Children absent at endline are quietly removed, which inflates results. A level becomes a permanent label and groups stop moving. And automated scoring is accepted without a facilitator check. Each one is cheaper to prevent in the data model than to explain later in a funder report.
How does FLN tracking differ for a small programme and a multi-state one?
The levels, tools and matched-reporting rule should be identical. A programme with a handful of centres can run on a structured sheet or a simple form with one assessor team. Across several states, expect different languages, school calendars and tool versions, so version mapping, offline entry, role-based access and a central back-check sample become essential rather than optional.